v0.5.3

Release Summary

BatchNorm Fixes & Improvements

• Fixed critical bias bug: Corrected line 59 in batchnorm.py where bias was incorrectly set to scale instead of B
• Fixed inference mode for batch_size > 1: Explicitly set BatchNorm to eval mode to use running statistics (ONNX inference behavior) #44
• Removed experimental flag: BatchNorm now works correctly with batch_size > 1 by default #35
• Added comprehensive tests: Validated against onnxruntime with various batch sizes (1, 2, 4, 8), channels, spatial dimensions, epsilon values, and momentum values

ReduceSumSquare

• Implemented ReduceSumSquare operator: Computes sum(x^2) along specified axes
• Supports both opset versions: Handles axes as attribute (opset < 13) and as optional input (opset >= 13)
• Comprehensive test coverage: 16+ parametrized test cases validating against onnxruntime with different input shapes, axes, and keepdims settings

LogSoftmax

• Added LogSoftmax operator: Supports axis attribute with proper default handling (dim=-1) #49 #30
• Added comprehensive tests: Each operator validated against onnxruntime with various input shapes, axes, and mathematical property verification

v0.5.2

Release Summary

This release adds significant new operator support and includes several important bug fixes.

New Operators & Features

• LayerNorm: Full LayerNorm operator support with comprehensive tests (#74)
• GRU: Added GRU (Gated Recurrent Unit) operator (#70)
• LRN: Added Local Response Normalization operator (#53)
• AutoPad: New AutoPad operation with support for auto_pad=SAME_UPPER (#58)
• Control Flow: Added If operator for conditional execution (#63)
• Sequence Operations: Added Optional and SequenceConstruct operators
• RandomUniformLike: Added RandomUniformLike operator (#57)

Enhancements

• Added comprehensive support for all ONNX attributes
• Implemented ONNX to PyTorch dtypes mapping for better type conversion

Bug Fixes

• Fixed issue with MatMul when it's the last operation (not followed by Add) (#59)
• Removed warning for read-only tensors (#73)

Miscellaneous

• Updated .gitignore

---

This release significantly expands ONNX operator coverage, particularly for sequence operations, control flow, and recurrent neural networks, while improving stability and type handling.

Contributors:

• gcunhase: #74

v0.5.1

• Added new operators
  • Hardsigmoid #65
  • Hardswish #60

v0.5.0

• Updated the library to work with latest onnx and pytorch versions
• Added new operators
  • GreaterOrEqual #56
  • ReduceL2 #54
  • ReduceMax #34 #33
• Switch to GitHub actions for CI

Special thanks to the contributors!

• https://github.com/zhumakhan
• https://github.com/calvinmccarter-at-lightmatter
• https://github.com/quaternio

v0.4.1

• Fix compatibility with torch=1.10.0
• Fix Clip operator when min and max are None
• Fix from onnx2pytorch import __version__

v0.4.0

• Improve memory management by removing activations that are no longer required by any following operations
• New supported operators
  • Tile
  • Loop
  • Bitshift
  • Div
  • Constant
  • GatherND
  • GlobalAveragePool
  • LSTM
  • Matmul
  • NonMaxSuppression
  • Prelu
  • ReduceSum
  • Scatter
  • ScatterElements
  • ScatterND
  • ThresholdedReLu
  • TopK
  • Transpose
  • Where

Main contributor for this release was @calvinmccarter-at-lightmatter

v0.3.0

• MLPerf v0.7 model support
• New operators support
  • Elu
  • And
  • Or
  • Not
  • Range
  • Expand
  • Unsqueeze - version 13
  • Squeeze - version 13

Special thanks to our first contributor @calvinmccarter-at-lightmatter for adding his part to this release.